The first group of people using AI have already developed AI fatigue.

Text | Silicon Research Laboratory Kiki

I wonder if you’ve noticed that many people are suffering from AI fatigue.

From the grand Token-factory plans of tech giants to the daily Token call volume refreshed every few days, we’ve entered an era of Token explosion—if you don’t burn through hundreds of thousands of Tokens and don’t have your own Skill, you can’t be called someone who understands AI.

In the past few days, whether it’s colleagues’ Skill and all kinds of Skills from former colleagues going viral, or the rise of another kind of anti-distillation Skill—everyone is trying to prevent their own knowledge and experience from being packaged into Skills by colleagues or bosses.

Two types of people are using the same set of tools—sounds pretty abstract and pretty absurd, doesn’t it?

But it truly reflects the anxiety and fatigue of ordinary office workers facing AI: AI is becoming more advanced and Token consumption is increasing. If AI can finish what we do in minutes in a day, why are we feeling more tired and more anxious instead?

1、When Token becomes the fourth salary

I know that many companies now have started incorporating AI usage into real work performance evaluations.

In many Chinese internet giants, performance reviews are directly linked to Tokens. Some departments have even built internal AI leaderboards—metrics including Token usage, AI code output rate, number of lines of AI code, and more. The more Tokens someone consumes, the higher their performance.

A product manager at a big company also told me that internally, they indeed encourage everyone to develop Skills, and there are weekly colleague showcases: “If you don’t develop, you’ll fall behind.”

Even more aggressive than China’s big tech companies might be their overseas peers. Meta formed an artificial intelligence team. At the earliest stage, it gamified AI usage with leaderboards; Google has also started to forcibly require some non-technical management personnel to use AI assistant Agents.

Inside JPMorgan Chase, they built a dashboard that tracks AI tool usage, and AI will tag employees: are you a light user, a heavy user, or a non-user?

Tech companies have started packaging Token quotas as “invisible benefits.”

In the past, big-company benefits depended on salary, free meals, and security measures. Now it’s about how many Tokens they can give you.

Alibaba plans to provide employees with Token quotas. Tencent provides employees with up to 220,000 Tokens per year. Nvidia is also preparing a Token budget for technical engineers equivalent to about half of their base salary. Huang Renxun even said:

“Token is the fourth form of compensation beyond salary, bonuses, and equity.”

When bosses use “carrots and sticks” to motivate ordinary office workers, for some people the shift from anxiety about performance scoring to anxiety about Token consumption marks the start of a new digital quantity-pushing game.

Whoever consumes more Tokens seems to represent higher work efficiency; whoever can write better Skills seems to show a deeper understanding of the business. An e-commerce industry practitioner told me that now companies have formed an invisible hierarchy: if no one is raising shrimp and the Token consumption isn’t enough, they’ll be looked down on.

So the question is: is this kind of evaluation system truly flawless?

The answer is obviously no.

2、Human context windows are running out fast

Why is the answer no?

Let’s first talk about a magical story from the media industry.

In some co-written submissions, the client usually provides a Brief. Lately, colleagues’ lived experience is that more and more Briefs are written by AI. Some clients certainly won’t say the Briefs are written by AI (even though the “DS vibe” is very strong), and some are more direct: the reason they give is, “There’s too much information—I used AI to generate a thought process for your reference.”

On the other hand, some clients also use tools that detect AI-written text to check the AI content in articles. Some articles, because the AI flavor is too strong, even spark discussions on social media.

Giving AI Briefs and writing AI articles creates a magical closed loop. But does it really make sense? When everyone raises shrimp and everywhere are Skills, does it truly bring a multiplier increase in productivity?

I also threw these questions to a circle of people around me in different industries—people who use AI in daily life.

A programmer told me that in her day-to-day work, 90% of the code is written with AI. The delivery timeline for one project has shrunk from one year to 4 months, but the work has been getting more and more, and project pressure has been increasing as well. This year, the small team she’s on has already voluntarily had two employees resign.

An algorithm engineer, also a heavy user of Vibe Coding, told me that now every day he doesn’t have time to respond to his Clade Code Session. He has multiple tasks running at the same time, and he feels his attention is seriously scattered. Sometimes he even forgets why he started in the first place.

There’s also a non-technical practitioner working in e-commerce operations. Her boss requires everyone to adopt AI and “raise shrimp.” It seems that now, without AI involvement, it’s not possible for copywriting, scripts, or product images.

A well-known AI blogger, Zhang Zala, had a post recently that made me feel a lot of resonance. She said that after deeply using AI, she is now in a “semi-ADHD” state. The AI context window (Context Window) is no longer the bottleneck, but human context windows are running out fast.

AI is evolving too quickly. Now carbon-based organisms are chasing silicon-based ones.

These AI anxieties and fatigue are what researchers call AI brain fry.

《Harvard Business Review》 conducted a research survey of 1488 U.S. full-time employees from large companies across various industries. The findings show that a considerable number of employees report symptoms such as sluggish thinking, headaches, and slower decision-making.

Why does AI brain fry happen? This study said three very interesting observations:

First, when using AI, the most mentally demanding part is supervising the AI. The study pointed out that high-level AI supervision also predicts additional mental fatigue for participants.

Second, AI increases workload. In addition to supervising AI, once AI is involved, it expands employees’ scope of responsibilities, requiring them within the same time not only to pay attention to more tools, but also to more results—greatly increasing cognitive load.

Third, using more and more AI tools doesn’t necessarily mean higher productivity. The study found that when employees move from using one AI tool to using three, productivity increases significantly. But after using three tools, productivity scores decline.

And the fundamental reason these kinds of fatigue arise is that we ignore the most valuable resource in human beings—attention scarcity.

3、AI is intensifying the attention crisis

If you ask those AI entrepreneurs and AI users how AI has changed their work, most of them will give you an optimistic answer:

AI has taken over more of my work, my workday has become easier, and improvements in efficiency are starting to show.

But what’s reality like? Most people are probably experiencing the AI fatigue and AI brain fry we mentioned earlier: embracing AI, but your work doesn’t decrease the way you imagined.

ActivTrak is a U.S. SaaS provider focusing on workforce analytics and productivity management. They recently conducted an interesting survey.

By collecting behavioral data from 1111 companies, 163638 employees, and more than 4.43亿 hours between January 1, 2023 and December 31, 2025, they found:

AI hasn’t redistributed the workload; it has increased work burden. The speed of the expansion of collaboration exceeds what people’s attention can handle. Yes, productivity improvements exist, but these improvements are increasingly dependent on fragmentation rather than deep focus.

A set of data in the survey is also shocking: workdays are shorter, but work hours start earlier, collaboration time increases, and attention is diluted.

• Focus efficiency drops to 60%, the lowest level in three years (63% in 2023)

• Average focus duration drops by 9%—from 14 minutes 23 seconds per day to 13 minutes 7 seconds.

• Collaboration volume surges 34%, reaching 52 million per day.

• Time spent on multitasking increases by 12%, reaching 1 hour 33 minutes per day.

• Weekend workload increases by more than 40%

This points to a hidden crisis behind AI progress: AI is intensifying an attention crisis. AI solves efficiency problems, but it also triggers deeper burnout.

The report also notes a phenomenon: the proportion of employees at risk of burnout increased by 23%, jumping sharply from 19% to 23%.

The reasons for burnout aren’t hard to understand either, because the energy that’s released isn’t managed effectively. Either bosses keep assigning work with higher intensity but not higher value, bringing greater work pressure, or individuals lack the ability to carefully plan their attention.

This is a key question many of us are facing right now: AI has saved visible time—so how should we use the time saved?

That becomes a new question, doesn’t it?

4、Written at the end

You probably didn’t expect that more than 100 years ago, Keynes predicted this problem.

This economist predicted that by the early 21st century, capital accumulation, productivity improvements, and technological progress should bring us to a stretch of “economic paradise,” where people’s weekly working hours wouldn’t exceed 15 hours. But he also asked:

How will people spend that leisure time?

More than 100 years later, even with technology being even more advanced, we still can’t give an answer.

AI, which brings enormous productivity improvements, also brings a new crisis of meaning and value. Workers can see the efficiency gains from AI tools, but they also feel busier, under more pressure, or find it harder to fully detach from work.

When consuming Tokens and creating Skills become new KPIs, workers scramble to push volume, trying to prove they haven’t been eliminated in the AI era—but in doing so, they accelerate the replacement of themselves by AI.

More critically, Keynes’s leisure time still won’t arrive in the short term. A harsh reality is: the few hours of free time you save through AI will be filled by more problems that need solving and more projects that need follow-up.

At least for now, AI fatigue won’t weaken or disappear, because AI is still staging even more aggressive changes—in fact, as I’m writing this, a new round of mass layoffs in Silicon Valley has begun.

Does AI bring work liberation, or is it a treadmill that speeds up the pace? This is gradually becoming the decisive question of this decade.

Reference materials:

1、Harvard Business Review: When Using AI Leads to “Brain Fry”

2、ActivTrak：2026 Stateof the Workplace——AI Adoption & Workforce Performance Benchmarks

A flood of information and precise insights—only on the Sina Finance APP

Responsible editor: Liu Wanli SF014